Automatic Threshold Derivation for Robot Vision Edge Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for setting threshold values in image feature detection are inconsistent and require manual tuning, failing to accurately account for image sensor noise, especially shot noise proportional to pixel brightness, which affects edge detection accuracy in robot assembly tasks.
Innovation Solution
An information processing apparatus that measures the relationship between luminance values and noise variation, using an image sensor noise model to predict luminance gradient noise, allowing for automatic setting of threshold values to differentiate between noise and actual edge features in images captured by cameras.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual tuning is used to set threshold values, then the threshold can be adjusted based on practical experience, but the process becomes inconsistent between different workers and requires significant manual effort
Solution Approach 1:
The system performs self-calibration by automatically capturing images of a calibration board, detecting features, and deriving threshold values without human intervention. The threshold value derivation unit automatically processes the captured images and computes optimal threshold values based on detected feature distributions, eliminating the need for manual tuning while ensuring consistency across different operating conditions.
Solution Approach 2:
The system performs preliminary calibration by capturing images of a known calibration board before actual measurement tasks. This preliminary action establishes reference data about the specific camera-imaging unit combination, allowing the system to pre-determine appropriate threshold values that account for individual device characteristics before they are needed for actual feature detection.
2Measurement precision
If threshold values are set to detect all luminance variations, then more features are detected, but noise features are also detected reducing measurement accuracy
Solution Approach 1:
The system applies different threshold values to different regions of the image based on local luminance characteristics. The threshold value derivation unit analyzes the luminance distribution in specific regions and derives region-appropriate threshold values, allowing the system to maintain high sensitivity in low-luminance areas while effectively rejecting noise in high-luminance areas where noise is more prominent.
Solution Approach 2:
The system dynamically adjusts threshold values based on the detected luminance gradient distribution. Instead of using a fixed threshold, the threshold value derivation unit computes optimal threshold values by analyzing the statistical distribution of luminance gradients in the captured image, adapting the threshold parameter to match the specific noise characteristics of each imaging condition.
3Measurement precision
If the imaging unit is fixed to a specific camera, then the threshold value setting can be optimized for that device, but the system lacks flexibility when using different imaging units
Solution Approach 1:
The system achieves universality by implementing a standardized calibration procedure that can be applied to any camera-imaging unit combination. The threshold value derivation unit is designed to work with different imaging units by performing the same calibration process - capturing images of a calibration board and deriving thresholds from the detected features - ensuring that the system adapts to each device's characteristics while maintaining a unified approach across multiple devices.
Data Source
AI summary
An information processing apparatus includes an image acquisition unit, variation amount deriving unit, acquisition unit, and threshold value deriving unit. The image acquisition unit acquires a captured image obtained by an imaging unit. The variation amount deriving unit derives a luminance value variation amount of a predetermined area of the acquired captured image based on a luminance value of the predetermined area and first information indicating a relationship between a captured image luminance value and an amount of luminance value variation. The acquisition unit acquires an amount of variation of a luminance gradient value based on the derived amount of variation of the luminance value, wherein the luminance gradient value is a gradient value of the luminance value. The threshold value deriving unit derives a threshold value with which an obtained luminance gradient value is to be compared, based on the acquired amount of variation of the luminance gradient value.